1 / 1

Machine Learning of Bridge Bidding By Dan Emmons Computer Systems Laboratory 2008-2009

Machine Learning of Bridge Bidding By Dan Emmons Computer Systems Laboratory 2008-2009. Performance

race
Download Presentation

Machine Learning of Bridge Bidding By Dan Emmons Computer Systems Laboratory 2008-2009

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Machine Learning of Bridge Bidding By Dan Emmons Computer Systems Laboratory 2008-2009 Performance Here are sample hands that the program bid. In the first one, the program uses nothing other than its expected value algorithm with Monte Carlo simulation. The program earns a good score, but this is in part due to luck. In the second one the program combined these techniques with the use of bidding conventions to more accurately express its hand and earns a fantastic score that is very much grounded in skill. The Goal The goal of this project is to create an effective machine bidder in the card game of bridge. Factors like partial information and the multiplicity of the meanings of bids make this task difficult. This research proposes to overcome these problems with the use of a Monte Carlo simulation method for overcoming the limitation of partial information and a tree structure of constraints paired with sets of actions to store the bidding system used by a partnership. With this tree structure a machine partnership trains by continually swapping their new bidding inclinations to learn new decision networks. The performance of bidders is evaluated by having them play against a control pair in both directions for each hand and converting the results to an average IMP gain per hand. The results of this project will not only demonstrate the feasibility of having a machine learn to bid in this manner, but also may develop new bidding conventions useful to human bridge players. Dealer: West Vulnerable: None North Clubs: A K 7 6 Diamonds: J T 8 4 Hearts: Q T 8 3 Spades: 2 West East Clubs: 9 8 5 4 Clubs: J 2 Diamonds: 9 7 6 Diamonds: A Q 2 Hearts: J 2 Hearts: A K 9 7 6 4 Spades: 8 7 6 3 Spades: K 9 South Clubs: Q T 3 Diamonds: K 5 3 Hearts: 5 Spades: A Q J T 5 4 South West North East Pass Pass Pass 2S Pass 3H Pass 3S X 4C Pass 4S Pass 4NT Pass 5C Pass 5H Pass 5S X Pass Pass Pass 5SX Nonvul - South Making Exact Score: 650 Development Monte Carlo simulation is used to allow bidders to envision other hands. Approximations of other hands become successively better throughout the auction because more bids give more information. Dealer: West Vulnerable: E-W North Clubs: Q J 8 4 3 Diamonds: K T 4 Hearts: K J 9 Spades: K 7 West East Clubs: T 6 2 Clubs: 5 Diamonds: A 5 3 2 Diamonds: 9 8 7 6 Hearts: 8 6 2 Hearts: Q 5 3 Spades: Q 6 2 Spades: A J 9 4 3 South Clubs: A K 9 7 Diamonds: Q J Hearts: A T 7 4 Spades: T 8 5 West North East South Pass 1D Pass 1H Pass 2C Pass 3NT Pass 4C Pass 4H Pass Pass Pass 4H Nonvul - South Making Exact 420 The bidding hierarchy used by the agents to determine what a certain conventional bid means or implies about their hand is represented as a tree, where each node contains hand constraints and actions. There are many priorities of pointers, and feasible actions are determined by taking the union of the sets of actions of the bottom-most nodes reached. Results The implemented bidding agents have performed spectacularly. Without the use of bidding conventions, their bid sequences are awkward and unpredictable, but the agents frequently score well regardless. Using these conventions the agents bid quite skillfully, both against other computers and human opponents. This has been quantitatively verified with IMP scoring matches against players of both types.

More Related